Boundary Sweep

When a seemingly self-contained system keeps surprising you, add one upstream layer, one downstream layer, one adjacent actor, and one externality, then prune what changes nothing.

The reporting dashboard your team owns breaks in a new way every month, and every fix is sensible, yet the surprises keep coming. When a system you've studied thoroughly keeps surprising you, the surprises usually aren't coming from inside the system. They're coming from just outside the box you drew around it, from places your analysis structurally cannot see.

Boundary Sweep is a 30-second expansion for exactly this cue: a seemingly self-contained system that repeatedly surprises you. The procedure has four additions and a pruning. Add one upstream layer: what feeds the dashboard (the data warehouse whose schema another team changes without notice). Add one downstream layer: who consumes it (the executive whose new filtering exposed old bugs). Add one adjacent actor: the sibling team whose jobs share your compute. Add one externality: the cost or effect landing elsewhere (your queries slowing their pipeline). Then prune ruthlessly: keep only the additions that actually change a prediction or explain a surprise, and drop the rest. It works when last month's mystery failure suddenly has an address, usually upstream. The spec's warning marks the stop rule: halt before "everything affects everything." A boundary that includes the whole company explains nothing and assigns no next action.